Skip to content

Auto Blemish

Detects the faces in a portrait and heals pimples, spots and small marks on the skin. The skin texture continues through each healed spot, so it does not look smoothed.

Auto Blemish — before, after, healed spots
In ComfyUI Add Node → ComfyUI-JH-PixelPro → face → Face: Auto Blemish
Good for A first pass of spot healing on portraits, group photos, batches of headshots, body skin with your own mask
Speed About 5 s on a CPU for an 8 MP portrait, under 1.1 GB VRAM

When to use it

This is the automatic version of the Spot Healing Brush. It does the dull part of portrait retouching: finding every small mark. You then check the result and fix what remains by hand.

Warning

It also removes natural moles unless you protect them. Draw a mask over any beauty mark, freckle pattern or tattoo you want to keep and connect it to protect_mask.

Each face is processed in its own crop, with its own spot size, so group photos work. Eyes, brows and the mouth are never touched. For skin without a face in the frame (back, hands, arms), give the node a mask of the skin area instead.

Quick start

  1. Load Image → Face: Auto Blemish → Preview Image.
  2. Keep the defaults. Raise sensitivity if faint marks are left. Lower it if freckles start to disappear.
  3. Connect spot_mask to Mask Preview to see exactly what was healed.
  4. Check face_count. If it is 0 on a portrait, the face was not found; use mask instead.

Settings

Setting Default What it does Tip
image The portrait to retouch.
sensitivity 0.5 0 to 1. Higher catches fainter marks. Very high also removes freckles. 0.3 to 0.6 for most portraits.
spot_scale 1.0 0.25 to 4. The largest mark to remove, relative to the automatic size (5 % of each face width). Raise it for bigger pimples. Lower it to keep small features.
mask (optional) Skin area to clean instead of the detected faces. Spot size then follows the mask width. For body, back and hands.
protect_mask (optional) Areas that are never touched. A beauty mark you keep, tattoos.

Outputs

Output What it is
image The retouched image.
spot_mask The pixels that were healed. Check it with Mask Preview.
face_count How many faces were found. 0 when a mask is given.

Tips from retouching

  • Always look at spot_mask once. It shows whether the node caught real blemishes or something you want to keep.
  • Natural moles and strong freckles are at risk at high sensitivity. Protect them with protect_mask.
  • Follow with Skin Smoother or Frequency Separation for tone evening. Auto Blemish only removes marks.
  • A mark that survived is often bigger than the automatic limit. Raise spot_scale, or heal it with Heal / Content-Aware Fill.
  • Large images work well: each face is cropped, so cost depends on the faces, not the whole frame.
Under the hood
  1. Region. Face Detect Lite (YuNet) finds the faces. The weights are downloaded once by Kornia. Each face box, grown by 25 %, is processed as its own crop. Only skin-coloured pixels are kept (a soft YCrCb test that covers light to dark skin), minus discs around the eyes, brows and mouth placed from the five landmarks. A user mask replaces the face search.
  2. Detection. The spot detector of Dust & Spot Removal runs on a signal made of luma darkening plus redness, because pimples are redder more than they are darker. The largest spot follows the face size (about 5 % of the face width).
  3. Healing. The texture fill of Dust & Spot copies the best nearby skin patch, matched on the ring and brightness-corrected, so pores continue.

Source: core/blemish.py, wrapper nodes/blemish_node.py.

Example workflow

Auto Blemish example workflow runs the sample portrait through Auto Blemish and shows the previews plus a Mask Preview of spot_mask. Copy workflows/sample_portrait.jpg into ComfyUI/input/ (Pexels License).